Pricing

The end of starting over.

Nimgir turns fragmented chats into one continuous, peak-quality conversation.

Desktop

For Claude Desktop users. Structured session continuity with quality awareness built in.

Learn more →

Terminal

For Claude Code users. Configurable quality thresholds and professional workflow tools.

Learn more →
What's next

Headless & fleets — Nimgir for automated runs and long-running agent fleets. Designed; announced here when finished.

Session report card — a shareable end-of-session quality card with a streak.

Coming to other agentic platforms.

Same engine, different workflows. Pick the one that fits how you work.

All products

Your AI doesn't have memory problems.
It was never designed to remember.

Every new session starts from zero. Your context, your decisions, your reasoning flow — gone. Not because something broke. Because that's how transformers work.

The Facts
Every AI model loses quality as conversations grow longer. This is proven across 18 frontier models. see research →
Your AI doesn't know it's degrading. It delivers confident, worse answers with no warning. see research →
Instructions get forgotten mid-session. The longer the conversation, the more gets skipped. see research →

How Nimgir works

WATCH — Nimgir keeps an eye on your session's health while you work: green, amber, red — always visible.

SAVE IN TIME — before quality drops, the session is wrapped up properly: what you decided, why, what's left, and exactly where you stopped.

CONTINUE — your next chat picks up like the same conversation. No re-explaining, no starting over.

SEE EVERYTHING — a page in your browser shows what your AI did and what needs you, in plain words.

One project. One mind. Every session.

What Nimgir Does

Nimgir makes every session a continuation of the last one — not just the facts, but the flow.

Continuation

Seamless Continuation

Mid-brainstorm, mid-code, mid-plan — the next session picks up your exact train of thought. Same reasoning style, same direction, same voice. Like talking to the same colleague every day instead of briefing a new one.

Persistence

Decisions That Stay Decided

What you locked stays locked. An append-only log carries your conclusions forward so nothing gets relitigated next session.

Awareness

Degradation-Aware Sessions

Long chats quietly lose quality — research proves it, and the AI never warns you. Nimgir keeps every session inside tested safe zones and wraps it up before quality drops, so you're never working with a tired AI without knowing it.

Continuity

Cross-Interface

Desktop and Terminal, same workspaces. Brainstorm in the browser, build in the terminal, nothing lost between them.

Already using the free version? The full version doesn't just continue — it remembers why.

Your work is safe by design

Not promises — how the software is built.

Everything stays on your computer

Your sessions and project memory are plain files on your own machine — you can open and read every one of them. Nothing is uploaded anywhere, except checking your license.

Your data can never be held hostage

The license system can only ever decline to manage NEW sessions. It cannot touch, lock, or delete anything of yours. If we disappeared tomorrow, all your work stays on your disk, readable.

If something breaks, your work doesn't

Internet down? Our servers down? You keep working. Nimgir steps aside gracefully — it never stands between you and your AI.

Saves that protect themselves

Every save is backed up automatically, and saves that look wrong are refused. Your project's memory can't be silently overwritten.

Pricing

One-time purchase. No subscription.

Try everything free first — the demo is the full product.

Desktop
$20 $30
one-time
Founding price — first 3 months after launch. List price $30.
For Claude Desktop users. Structured continuation with session-managed quality awareness.
Includes 12 months of updates — the license itself never expires.

Need more control? See the Terminal version →

Prices are in US Dollars (USD). For other currencies, conversion is handled by your card-issuing bank.

Buy once, keep forever. Your license is perpetual — the software never locks, never expires, never phones home to ask permission. Purchases include 12 months of updates (new features and compatibility with new Claude releases). After that, the software keeps working exactly as it is; extending the update term is optional.

All products

You already know context is finite.
Now you can engineer around it.

Every token competes for the same fixed attention budget. As sessions grow, quality degrades — silently, confidently, invisibly. You've seen it. Nimgir gives you the architecture to manage it.

What the Research Shows
Effective context windows range from 100 to 2,500 tokens across 11 LLMs tested. Up to 99% of the advertised window is unusable for complex reasoning. Paulsen et al., 2025 — MECW see research →
Replacing all irrelevant tokens with whitespace still degraded performance. Input length itself damages reasoning — not content complexity. Du et al., 2025 see research →
When you use AI, you overestimate how well you're performing. Higher AI literacy makes it worse, not better. You have no reliable self-assessment signal. Aalto University, 2026 see research →
Observation masking outperforms LLM summarization for agent context management. Dumping raw context into the window makes things worse, not better. JetBrains / TU Munich, NeurIPS 2025 see research →
Anthropic's own research describes context as a finite resource with diminishing marginal returns — a limited attention budget every token depletes. Anthropic, 2025 see research →
All 18 frontier models tested — GPT-4.1, Claude 4, Gemini 2.5 — grow increasingly unreliable as input length increases. Universal, replicated, architectural. Chroma Research, 2025 see research →

The quality loop

MEASURE — reasoning volume and context fill, tracked live per session, with zone thresholds.

WRAP IN TIME — before the degradation zone: a structured briefing — plan state, every decision with its WHY, dead ends, discovered context, exact stop point.

RESUME MID-THOUGHT — the next session boots loaded: briefing, project knowledge, locked-decision log. Continuation state, not a summary.

OBSERVE — a live status page: agent activity, session health, what needs the human. Zero terminal archaeology.

Session Infrastructure, Not a Wrapper

A session quality manager built as an MCP server — context managed as the finite resource it is. No prompt templates. No magic. Architecture.

Handover

Continuation Briefings

Structured handover documents carry reasoning chains, decision context, and exact stop points across sessions. Not summaries — continuation state.

Persistence

Append-Only Decision Log

Locked conclusions persist across sessions without re-derivation. The log is the source of truth — no drift, no relitigating.

Threshold

Configurable Quality Cap

Calibrated quality caps out of the box — derived from measured field data, per platform — with a zone ramp that warns, then wraps. Tune the thresholds to your own workflow when you're ready; the defaults already know where the danger zone is.

Workspace

Isolated Workspaces

Independent context per project. No cross-contamination. Same workspace accessible from Desktop and Terminal — brainstorm in browser, build in CLI.

Already using the free version? The full version doesn't just continue — it remembers why.

Safety properties, not policies

Local-first by architecture

Briefings, logs, knowledge — plain text on your disk. No cloud copy of your work. The only network call is license validation.

The never-ransom invariant

The license layer can decline to manage NEW sessions — that is its entire authority. Touching, locking, or deleting user data is structurally outside its code path. Server death leaves every byte readable.

Fail-open for licensed users

Network or server failure degrades session management, never your session. No network calls on the hot path of your work.

Guarded, atomic saves

Staged writes, atomic commits, automatic backups on every save, and guards that refuse suspicious shrinks and overwrites. Designed against a hard-won lesson: sometimes the writer you defend against is a degraded AI.

Pricing

One-time purchase. No subscription.

Try everything free first — the demo is the full product.

Terminal
$100 $150
one-time
Founding price — first 3 months after launch. List price $150.
For Claude Code users. Includes the full Desktop version, plus configurable quality cap and professional workflow tools.
One license covers 2 machines.
Includes 12 months of updates — the license itself never expires.
Terminal Team Bundles
Team 5 — 5 people × 2 machines each
$600
Team 10 — 10 people × 2 machines each
$1,000
Team 20 — 20 people × 2 machines each
$1,800
Every Terminal seat includes the full Desktop version.

Prices are in US Dollars (USD). For other currencies, conversion is handled by your card-issuing bank.

Buy once, keep forever. Your license is perpetual — the software never locks, never expires, never phones home to ask permission. Purchases include 12 months of updates (new features and compatibility with new Claude releases). After that, the software keeps working exactly as it is; extending the update term is optional.

The Research Behind Nimgir

Every claim on this site is backed by published research. No marketing assertions. Here are the sources.

Evidence grades classify each finding's strength: Established means scientific consensus across multiple independent studies. Supported means strong evidence with some debate on scope. Emerging means early evidence that needs further validation.

Where Nimgir comes from

Nimgir wasn't designed on a whiteboard. It came out of 1,300+ real working sessions — including projects that failed BECAUSE sessions degraded silently. We measured the degradation signature (shallow answers, then drift, then confident fabrication), mapped how every summary-handoff bleeds reasoning like a telephone game, and calibrated the caps on field data. The studies below say the same thing from the outside: this is architectural — every long-session AI, not one vendor. Nimgir exists so you can build ON these models at their best.

Emerging Paulsen et al., 2025

Maximum Effective Context Window (MECW)

Across eleven LLMs, the effective context window — the amount of context the model can actually use for complex reasoning — ranges from as low as 100 to 2,500 tokens depending on model and task type. Up to 99% of the advertised context window may be unusable for complex tasks.

Read the research
Established Du et al., 2025

Length Itself Causes Degradation

In a controlled experiment, all non-relevant tokens were replaced with blank spaces — and performance still degraded. This proves that input length itself damages reasoning, not content complexity. The problem is architectural, not solvable by better prompting.

Read the research
Supported Aalto University, Finland

The Reverse Dunning-Kruger Effect

When humans use AI, they consistently overestimate their own performance. Higher AI literacy makes this worse — more technical knowledge brings more confidence, not more accuracy. The traditional Dunning-Kruger effect disappears: everyone overestimates, regardless of skill level. You think you're doing better than you are.

Read the research
Established JetBrains Research & TU Munich — NeurIPS 2025

The Complexity Trap

Observation masking outperforms LLM summarization for agent context management. Raw file contents and exploration threads should be extracted for decisions and status markers, not fed whole into context. More context does not mean better performance — it often means worse.

Read the research
Supported Anthropic, 2025

Context as a Finite Resource

Anthropic's own research describes context as "a finite resource with diminishing marginal returns" — a limited attention budget that every new token depletes. The maker of Claude acknowledges the architectural limitation their product operates under.

Read the research
Established Chroma Research, 2025

Multi-Model Context Degradation Study

Tested 18 frontier models including GPT-4.1, Claude 4, and Gemini 2.5. Every single model's performance grows increasingly unreliable as input length grows. Performance degradation as context length increases is real, universal, and replicated.

Read the research

Legal Documents

Download Nimgir

Same installer for demo and full version. Your license key unlocks everything.

Try the full product — not a crippled version

The demo is Nimgir complete: 5 chats on Desktop, 10 working sessions on Terminal — enough to live through several wrap-and-resume cycles and feel what never re-explaining your project means. Your session counter is visible from the first minute, and only sessions that complete a real wrap count. When the demo ends, your data stays on your disk, readable, yours — and the free session meter keeps working. One command to install. No signup, no card.

Desktop
For Claude Desktop on Windows and macOS
v1.0 · .bat
Download Installer

Installs the MCP server locally. No admin rights needed. ~120 KB.

Terminal
For Claude Code on Windows and macOS
Coming Soon

Who Nimgir v1.0 is for

Built for people working WITH Claude: Claude Code in a terminal — single or multiple terminals; Claude Desktop; Windows and macOS.

Not yet for: headless and automated runs (claude -p, CI pipelines, Agent SDK fleets) — designed, on the roadmap, announced here when finished. Linux arrives with that track.

What Nimgir doesn't do

It doesn't make the model smarter, and it can't rescue a session that has already degraded. It makes sure you stop BEFORE that point — with everything intact — and that the next session starts at peak with the full story. Prevention, not cure.

Stop re-explaining. Start continuing.

Setup Guide

Step-by-step walkthrough to get Nimgir running in your Claude environment.

Interactive Setup Wizard

Illustrated guide with copy-paste blocks, platform detection, and license activation. Works for both demo and paid users.

Open Setup Guide →

Documentation

Day-to-day usage guide for Nimgir after installation.

How Nimgir Works

Claude doesn't have memory between conversations. Every new chat starts from zero — your context, decisions, and reasoning chain are gone. This isn't a bug; it's how transformer architecture works.

Nimgir bridges that gap. It's an MCP server that runs locally on your machine and gives Claude structured handover documents at the start of every chat. Instead of starting fresh, Claude picks up exactly where the previous session left off.

The Three Layers

Primer — A fixed set of rules that ship with Nimgir. These tell Claude how to manage sessions: when to wrap up, how to format handover documents, how to track token usage. You never edit the primer — it loads automatically on every get_briefing call.

Briefing — A narrative continuation document written by Claude at the end of each session. It captures the reasoning chain, the exact stop point, open threads, and what the next session should do first. Think of it as a story of how the thinking evolved, not a list of bullet points.

Master Log — An append-only record of locked decisions. Only finalized conclusions go here, tagged with persistent markers that survive across archive cycles. The master log is the single source of truth for what was decided.

Key principle: The briefing carries the narrative thread (what you're thinking about). The master log carries the locked decisions (what you've concluded). Together, they give the next Claude everything it needs to continue without re-explaining.

What Is a Workspace?

A workspace is an isolated container for one project or topic. Each workspace has its own briefing, master log, persistent tags, and archive history. Nothing leaks between workspaces — your web-app project and your research notes never cross-contaminate.

Naming Rules

Workspace names must be lowercase letters, digits, and underscores only, between 1 and 64 characters. Examples: my_project, research_2026, client_alpha.

Creating a Workspace

Workspaces are created automatically the first time you call save_briefing with a new workspace name. If the name isn't registered yet, Nimgir asks Claude to confirm with you before creating it — this catches typos before they become orphan folders.

Listing Workspaces

Ask Claude to call list_workspaces at any time. It returns all registered workspace names with their last-updated dates. Useful for remembering what you named things or checking when a project was last active.

Multiple Workspaces

You can have as many workspaces as you need. To switch between them, just start a new chat and use a different workspace name in your trigger text. Each project or topic gets its own clean context.

The Session Lifecycle

Every Nimgir session follows the same four-step cycle: Load → Work → Wrap → Continue.

1. Load

At the start of each chat, Claude calls get_briefing with your workspace name. Nimgir returns the primer (session rules), the briefing (narrative continuation from last time), and the master log (locked decisions). Claude reads all three and picks up from where you left off — no re-explaining needed.

2. Work

Normal conversation. Claude tracks token usage with a bar at the bottom of each response, showing how much of the quality budget has been used. You work on your project, make decisions, explore ideas — business as usual.

3. Wrap

When the token bar approaches the quality limit (around 55K tokens), Claude lets you know. By 58K, it begins wrapping: writing a new briefing that captures the reasoning chain, locking any new decisions into the master log, and preparing the handover for the next session.

Claude calls save_briefing with the updated briefing and master log. Nimgir writes both files to disk with automatic backups.

4. Continue

Start a new chat. Claude calls get_briefing again and reads the briefing you just saved. The cycle repeats. From your perspective, it's one continuous conversation — each segment runs inside its tested safe zone, at peak quality.

Why split sessions? Research shows all AI models lose quality as conversations get longer. Nimgir doesn't fight this — it works with it. Short, high-quality segments with structured handovers beat one long degrading session every time.

The Token Bar

Claude posts a token bar after every response, showing estimated usage against the quality budget. The bar starts at 35K (the base context load — primer, briefing, master log, and system prompt consume tokens before any conversation begins).

~42K/60K ████░░░░░░

The Zones

35K – 54K: Safe. Full quality. Work normally.

55K – 57K: Caution. Claude warns you the safe zone is ending and offers to begin wrapping. You can push a bit further if needed, but quality starts thinning.

58K – 60K: Danger. Claude begins the wrap protocol regardless. Past this point, even the handover itself starts losing quality — better to wrap now than risk a degraded continuation.

How Tokens Are Estimated

The rough rule is 4 characters ≈ 1 token. This includes everything in the context window: the system prompt, primer, briefing, master log, your messages, and Claude's responses. The estimation isn't exact, but it's reliable enough to keep sessions in the safe zone.

Tip: If your master log is large (lots of prior decisions), your safe working window shrinks because more tokens are consumed before you start. Archiving old decisions reclaims that space.

"Nimgir tools not found"

Claude can't see the Nimgir MCP server. Check that your Claude Desktop configuration includes the Nimgir server entry, and that the Python path and server script path are correct. Restart Claude Desktop after any config changes.

"Unknown workspace" Warning

You used a workspace name that isn't registered. This is usually a typo. Ask Claude to call list_workspaces to see what names exist. If you intended a new workspace, Claude will ask you to confirm the creation.

Quality Seems to Be Dropping

Check the token bar. If it's above 55K, you're past the safe zone. Start a new chat — the handover will carry your context forward at full quality.

Contact Support

For issues not covered here: support@nimgir.ai

I run Claude headless — scripts, CI, the Agent SDK. Can I use Nimgir?

Not yet — and we'd rather tell you that than sell you a half-fit. Nimgir v1.0 is built around interactive sessions, where its checkpoint and wrap discipline can do their full job. A headless edition (long-running agent fleets first, high-volume operations after) is designed and on the roadmap. If that's you, tell us what you run — it shapes what we build first: support@nimgir.ai

What happens when my demo ends?

Nimgir stops managing new sessions. Your data — briefings, logs, project knowledge — stays on your disk, readable, yours. The free session meter keeps working. Buy a license and the same install continues where it left off.

Do you see my work?

No. Your sessions, briefings, and project files never leave your machine. The license server processes exactly: your key, your email, a machine identifier, and the product version. That's the whole list — it's in our privacy policy, word for word.

What if Nimgir the company disappears?

The software keeps working. Licenses are perpetual, validation fails open for paying users, and your data is plain text on your own disk. The never-ransom law is built into the code: the license system can decline to manage new sessions, and that is ALL it can do.

Advanced Guide

Archiving, persistent tags, file locations, data recovery, and license management — included with your license key.

Order Summary

Product
Delivery Digital license key via email
Email
Payment Credit / Debit Card (3D Secure)
Total
← Back to product

Complete Your Purchase

Enter your email to receive your license key after payment.

v1.8